[Neil Schemenauer]
> I seem to remember someone saying that GCC generated better code for:
>
> if (exceptional) {
> do exceptional things
> break / return / goto
> }
> do normal things
>
> Is GCC in the dumb category?
Yes, any compiler that doesn't do branch prediction based on *semantic*
analysis is dirt dumb. A simple example of semantic prediction is
"comparing a pointer to NULL is probably going to yield false". Ditto
comparing a number for equality with 0.
I'd like to see a reference for the pattern above; it goes against the very
common "forward branches usually aren't taken" heuristic. Note that
Vladimir applied that gimmick to an extreme in obmalloc.c's malloc()
function.
> Also, the Linux is starting to use this set of macros more often:
["likely" and "unlikely"]
They're late to the party. Cray had a "percent true" directive 20 years
ago, allowing for 48 bits of precision in specifying how (un)likely <wink>.
> ...
> I don't have GCC >= 2.96 otherwise I would have tried sprinkling some of
> those macros in ceval and testing the effect.
Maybe more interesting: One of the folks at Zope Corp reported getting
significant speedups by using some gcc option that can feed real-life branch
histories back into the compiler. Less work and less error-prone than
guessing annotations.
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4